Fix StatsModels

[nignatiadis]

Hi! This should fix #10, although it might be more elegant to proceed as @kleinschmidt recommended by defining the two lines of logic in StatsModels and including here only:

function concrete_term(t::Term, xs::AbstractVector{<:EventTime}, ::Nothing)
    IdentityTerm(t.sym)
end

@kleinschmidt: Does it matter if we dispatch on concrete_term rather than schema?

[kleinschmidt]

I think this fixes the problem in the short term but I'd rather fix it upstream by not converting to Float64 as aggressively for continuous terms. I won't have time to look at that for a few weeks though so if you need a short-term fix this will be fine. Otherwise I'd suggest keeping statsmodels upper bounded until there's a fix there.

[kleinschmidt]

As for dispatching on concrete_term vs. schema, I think you've made the right choice here (and that confusion is why I renamed what used to be methods of schema to concrete_term instead...)

[nignatiadis]

Sounds great, thanks! The new formula system is really cool and what I had been missing most in Julia and ... one of these days (maybe in June) I will find the time to properly implement B/natural-splines..

[ararslan]

Superseded by #17